To avoid overlap with the training data:
The model was trained on games from the
Lichess public database
played between July 2024 and July 2025.
For an unbiased evaluation, avoid using games played
during this period when testing the model.
Rating Distribution:
The charts show the model's predicted probability distribution over the player's
rating based on the game up to the current move. Each bar represents a 100-point
rating range. For example, if the bar for 1500 is 0.12, the model assigns a
12% probability to the player's rating falling between 1500 and 1600.
Point Estimate:
A single-number estimate of the player's rating based on the game up to that move.